Trade and African Regional Agreements: a spatial econometric approach

Transcription

1 Trade and African Regional Agreements: a spatial econometric approach Wilfried Koch, a, Aligui Tientao b, and Diègo Legros b, a Université du Québec À Montréal, b Université Bourgogne FrancheComté First Draft: 18 december 2012 Abstract The purpose of this paper is to evaluate the impact of African regional blocs on African trade ows while allowing for spatial interdependence between trade ows. To this end, we derive a spatial gravity equation by removing the implicit assumption that trade ows between two trading partners are independent of what happens in the rest of the trading world. We estimate the border eects for ve Sub-Saharan regional blocs (cemac, comesa, ecowas, sadc and waemu). We decompose the border eect into two components: a trade-boosting intra-bloc eect and a trade-reducing inter-bloc eect. Our ndings show that trade agreements produce positive eects on intra-bloc trade ows and these eects are particularly prominent when the blocs are advanced in their integration process. In addition, the spatial interdependence between trade ows is reected in a negative relationship as implied by the theoretical model, suggesting a natural measure of spatial competition. KEYWORDS: Border eects; Regional blocs; Spatial Autocorrelation; Sub- Saharan Africa. JEL: F12; R12; O55. Université du Québec À Montréal, Dept. of Economics, ESG-UQAM and CIRPEE, Phone: (+1) (2568), koch.wilfried@uqam.ca Laboratoire d'économie de Dijon, Université Bourgogne FrancheComté, 2 boulevard Gabriel, BP 26611, DIJON CEDEX, Phone: +33 (0) , aligui.tientao@u-bourgogne.fr (corresponding author) Laboratoire d'économie de Dijon, Université Bourgogne FrancheComté, 2 boulevard Gabriel, BP 26611, DIJON CEDEX, Phone: +33 (0) , diego.legros@u-bourgogne.fr 1

2 1 Introduction For more than two decades now we have witnessed the proliferation of regional blocs among developing countries (Collier and Venables, 2009). Their impact on trade ows is generally seen as positive in most developing countries especially in Africa which is marginalized in world trade. Through these regional blocs, African countries hope to increase the size of their markets and to secure the welfare associated with increased trade. A number of authors have attempted to evaluate the eect of regional blocs on the trade ows of Sub-Saharan African countries. Foroutan and Pritchett (1993) compared actual trade with what a traditional gravity model would predict. They found that trade ows between African countries are not below expectations. The median Sub-Saharan African share of intra-trade averages 8.1% while the predicted value is just slightly lower at 7.5%. Carrère (2004) showed that African trade agreements have generated a signicant increase in trade among members. Musila (2005) reported positive eects for ecowas and comesa. According to Behar and Edward (2011), sadc countries trade with each other more than twice as much as other pairs do. This literature claims that the regional agreements in Africa have slightly increased intra-zone trade ows. Other authors argue, on the contrary, that regional agreements do not have a significant impact on trade ows. Longo and Sekkat (2004) showed that, besides traditional gravity variables, poor infrastructure, economic policy mismanagement, and internal political tensions have a negative impact on trade among African countries. Except for political tensions, the identied obstacles are specic to intra-african trade, since they have no impact on African trade with developed countries. Coulibaly and Fontagné (2006) analyzed the location of countries, whether they are landlocked or not, and the quality of their road infrastructures. They found that the lower the percentage of paved tracks between countries, the greater the impact of this infrastructure improvement on import ows. Geda and Kebret (2008), investigating the case of comesa, showed that regional blocs had an insignicant eect on bilateral trade ows. The performance of regional blocs is mainly constrained by problems of variation in initial conditions, compensation issues, real political commitment, overlapping membership, policy harmonization, lack of diversication and poor private sector participation (Geda and Kebret, 2008). Introducing into the gravity equation a variable that captures informal markets trade, Agbodji (2007) argued that the existence of these markets signicantly reduced formal trade across Sub-Saharan Africa. More recent works also highlight the poor quality of infrastructures to explain the low level of intra-african trade ows (Bosker and Garretsen, 2012; De-Sousa and Lochard, 2012). Several methods have been used to assess the impact of regional blocs, especially the gravity approach (Aitken, 1973; Sapir, 1981). Initially, there was no theoretical 2

3 foundation for the gravity equation. The rst theoretical development was given by Anderson (1979) and was based on constant elasticity of substitution (ces) utility. Other theoretical frameworks were developed to account for the gravity relationship in the 1980s (Bergstrand, 1985; Helpman, 1984). These authors took into account two key determinants characterizing new trade theory models: economies of scale combined with product dierentiation and transport costs. Baier and Bergstrand (2001) developed a gravity model based on monopolistic competition, whereas other approaches focused on Heckscher and Ohlin's model (Deardor, 1995; Evenett and Keller, 1998), or technological dierences between countries (Eaton and Kortum, 2002). The disturbing common feature of the previously mentioned studies resides in the implicit assumption that trade ows between two trading partners are independent of what happens to the rest of the trading world. This is clearly a very strong assumption that is unlikely to hold and, therefore, may lead to biased and inconsistent estimates of the gravity equation. Paul Krugman (1991) pointed this out and presented a model of trade between two regions. Following the publication of his paper, several studies have tried to model space and to demonstrate how space, via transport costs, can explain several puzzles in international economics. Anderson and van Wincoop (2003; 2004) show that the relevant export costs for exports from country i to country j, is the cost of exporting from i to country j relative to the cost of exporting from i to all other potential importing competitors of j. They demonstrate that after taking into account their size, trade ows between two regions decrease with their bilateral trade barrier, relative to the average barrier of the two regions in trade with their partners. The average trade barrier is called "multilateral resistance". Feenstra (2002) used an alternative method to control for multilateral resistance by including exporter and importer-specic xed eects. A further method was introduced by Baier and Bergstrand (2009) and applied to free trade agreements by Behar and Cirera-i-Crivillé (2013), which linearizes the Anderson and van Wincoop's system in such a way that multilateral resistance is captured by a linear function of observable trade costs. They use a Taylor-series expansion to solve for multilateral resistances. But this approach requires a normalization of the resistances to a reference country, so each computed multilateral resistance must be interpreted relative to a particular country that has to be chosen in advance. Kelejian et al. (2012) specify and estimate a generalization of the typical gravity model which includes country pair xed eects, third country eects, endogenous regressors, and error terms that are both spatially and time autocorrelated. Behrens et al. (2012) takes into account the spatial interdependence between trade ows a spatial gravity model to control for multilateral resistance. Furthermore, one notable characteristic of regional integration in Africa has been the multitude of regional integration initiatives and consequently the participation of 3

4 African countries in several of these regional trade agreements (RTAs). Many African countries hold multiple memberships. Of the 54 countries, 27 are members of two regional groupings, 18 belong to three, and one country is a member of four. Only seven countries have maintained membership of one bloc. There are some obvious limitations to this overly-complex regional integration architecture. Multiple arrangements and institutions, as well as overlapping membership in the same region, tend to confuse integration goals and lead to counterproductive competition between countries and institutions (ECA, 2008). Indeed, the RTAs pursue their own separate mandates and approaches to regional integration, imposing conicting requirements on countries that are members of more than one grouping and wasting scarce administrative and nancial resources. The overlapping of regional blocs shows that what happens in one bloc depends on what happens in another bloc. This means that there is a spatial interdependence between trade ows. And yet, this interdependence can be the source of spatial autocorrelation or spatial heterogeneity. While spatial heterogeneity can generally be treated by using standard econometric tools, the presence of spatial autocorrelation substantially changes the properties of estimators and the statistical inferences based on these estimators (LeSage and Pace, 2009). If there is spatial interdependence, the model of Anderson and van Wincoop (2003) and the subsequent studies are likely to yield biased estimators. Thereby, we use spatial econometrics tools to avoid this problem. The rst model that takes into account the spatial interdependence is that of Behrens et al. (2012). This model makes it possible not only to control for multilateral resistance but also to take into account spatial interdependence between trade ows. In what follows, we draw on Behrens et al. (2012) to derive a spatial gravity equation from the quantity-based version of ces. To the best of our knowledge, there are no analytical studies of African trade ows that take into account this interdependence. This paper aims to contribute to a wide literature on African regional agreements by assessing the eect of regional blocs on the trade ows of Sub-Saharan African countries. We put the main focus on spatial interdependence between trade ows by estimating the border eects for ve African regional blocs: Communauté Economique et Monétaire de l'afrique Centrale (cemac); Common Market for Eastern and Southern Africa (comesa); Economic Community of West African States (ecowas); Southern African Development Community (sadc) and West African Economic and Monetary Union (waemu) (see Table A in Appendix for the member countries of these blocs). These ve blocs are the main trade agreements in Africa (Carrère, 2004). All of the previously mentioned blocs dier in their degree of integration. We consider that waemu and cemac are closely integrated compared to other blocs, 1 although 1 Because they have managed to establish common external taris and they each have a common currency. 4

5 intra-bloc trade still experiences diculties (Goretti and Weisfeld, 2008; Martijn and Tsangarides, 2008). We expect a strong border eect for waemu and cemac. In terms of deeper integration, sadc is commonly viewed as the third most integrated bloc in Africa. Even if sadc countries do not form a customs union or do not have a common currency, they have nevertheless successfully implemented a free trade area (Behar and Edward, 2011). The aim of ecowas is to promote economic integration and cooperation with a view to creating an economic and monetary union for fostering economic growth and development in West Africa even if ecowas has not yet achieved its goals (Carrère, 2004; Musila, 2005). As regards comesa, it tries to achieve the removal of all physical, technical, scal and monetary barriers to intra-regional trade and commercial exchanges. However, like ecowas, comesa is struggling to achieve its goals (Geda and Kebret, 2008). Our methodology consists in removing the implicit assumption that trade ows between two trading partners are independent of what happens in the rest of the trading world. The basic idea is to get rid of prices and price indexes by using inverse demand functions and the fact that price indexes depend on trade ows. By doing so, we obtain a gravity equation that depends exclusively on observable variables and on a spatial autoregressive structure in trade ows. We decompose the border eect into two components: a trade-boosting intra-bloc eect and a trade-reducing inter-bloc eect. Our ndings show that trade agreements produce positive eects on intra-bloc trade ows and these eects are particularly prominent when the blocs are advanced in their integration process. With respect to the spatial eect, we nd a negative relationship between trade ows that can be interpreted as spatial competition. The remainder of the paper is organized as follows: Section 2 presents the theoretical model. In Section 3, we discuss our empirical results and Section 4 concludes the discussions. 2 The theoretical model We follow Behrens et al. (2012) by deriving a system of gravity equations that does not depend on unobservable price indexes, yet encapsulates the general equilibrium interdependencies of the full trading system. To this end, we build upon a ces trade model like those of Dixit and Stiglitz (1977) and Krugman (1980). More specically, we derive a gravity equation from the quantity-based version of the ces model by exploiting the fact that the price indexes are themselves implicit functions of trade ows. We obtain an implicit equation system that depends on observable variables only and that can be estimated using techniques borrowed from the spatial econometrics literature. 5

6 2.1 Consumers We consider an economy with n countries. Each country i is endowed with L i consumers/workers and each one supplies inelastically one unit of labour. Labour is the only production factor so that L i stands for both the size of, and the aggregate labor supply in country i. All consumers have identical and homothetic preferences over a continuum of horizontally dierentiated product varieties (Anderson and van Wincoop, 2003). A representative consumer in country j solves the following problem: max U j q ij (ν) σ 1 σ dν s.t. q ij (ν)p ij (ν)dν = y j (1) i Ω i i Ω i where σ > 1 denotes the constant elasticity of substitution between any two varieties; y j stands for individual income in country j; p ij (ν) and q ij (ν) denote the consumer price and per capita consumption of variety ν produced in country i; and Ω i denotes the set of varieties produced in country i. Since varieties produced in the same country are assumed to be symmetric, in what follows we alleviate the notation by dropping the variety index ν. Let m k stand for the measure of Ω k (i.e., the mass of varieties produced in country k). The aggregate inverse demand functions for each variety are given by: p ij = Q 1/σ ij k m kq 1 1/σ kj Y j (2) where Q ij L j q ij denotes the aggregate demand in country j for a variety produced in country i; and where Y j L j y j stands for the aggregate income in country j. 2.2 Firms It is assumed that the products are horizontally dierentiated and that each variety is produced by a single rm only. The production of each variety is subject to increasing returns with a common technology for all countries. Labour is the only factor of production, and in order to produce q units of output, cq + F units of labour are required, where c is the marginal cost and F the xed cost. Since shipping varieties both within and across countries is costly, shipping one unit of any variety between countries j and k requires dispatching τ jk > 1 units from the origin country j, so that p jk = τ jk p j, where p j is the mill price (Samuelson, 1952). A rm located in country j maximizes its prot, given by: π j = k (p jk cw j τ jk )Q jk F w j (3) Using equation (2) in the prot maximization process of the rm yields p j cw j σ/(σ 1). Free entry and exit drive prots to zero, which implies that each rm must produce the 6

7 break-even quantity τ jk Q jk = k F (σ 1) c Q (4) 2.3 Equilibrium To derive the gravity equation, it is necessary to know the value of trade ows from country i to country j at equilibrium. This is given by X ij m i p ij Q ij. Using equation (2), we obtain: X ij = m i Q 1 1/σ ij k m kq 1 1/σ kj From (5) we derive the following gravity equation 2 Y j (5) X ij = Y σ j [ k L k L i ( ) 1/σ 1 τkj Y k X 1 1/σ kj τ ij Y i ] σ i, j (6) which is a system of equations capturing the interdependence of all trade ows towards country j. To close the general equilibrium system, we impose the aggregate income constraints: Y i k X ik = 0, i (7) As can be seen from expressions (6) and (7), all equilibrium trade ows (including ows X ii ) are related directly (as the varieties of products are substitutes) or indirectly (through the national income). In the following section, we derive a spatial econometric reduced form by linearizing (6) to obtain an estimable equation taking into account all these interdependencies. 2.4 Econometric specication To obtain an econometric specication, we take equation (6) in logarithmic form: ln X ij = σ ln Y j σ ln [ k L k L i ( τkj Y k τ ij Y i ) 1/σ 1 X 1 1/σ kj ] (8) Clearly, there is interdependence across trade ows as X ij depends negatively on the nominal sales of the other countries in market j. To obtain a specication that can be 2 See Behrens et al. (2012) for the formal derivation. 7

8 estimated using spatial econometric techniques, we linearize (8) around σ = 1. Doing so yields the following equation: ( ln Z ij = σ ln L (σ 1) where Z ij X ij /(Y i Y j ) 3 ln τ ij k L k L ln τ kj ) σ ln w i (σ 1) k L k L ln Z kj (9) is a gdp-standardized trade ow (but which we will refer to as trade ow for short); and where L k L k denotes the total population. Expression (9) reveals the essence of spatial interdependence in the gravity equation: the trade ow X ij from country i to country j also depends on all the trade ows from the other countries k to country j. Several comments are in order. First, trade ows from i to j are aected by relative trade barriers, as measured by the deviation of bilateral trade barriers τ ij from the population weighted average (second term). Put dierently, relative accessibility matters. Second, trade ows from i to j are negatively aected by wages w i in the origin country (third term). Higher wages raise production costs and make country i's rms less competitive in market j, thereby reducing trade ows. Last, trade ows from i to j decrease with trade ows Z kj from any third country k into the destination market, because varieties are substitutes. This eect is stronger the closer substitutes the varieties are (i.e., the larger the value of σ). In our estimations, interdependence will be captured by an autoregressive interaction coecient, and this coecient can be seen as a measure of spatial competition encapsulating both aspects related to market power and consumer preference for diversity (via the parameter σ). As regards the functional form of trade costs, we assume that τ ij is a log-linear function of distance, border eect, landlocked position/status, common language, common currency and an error 4 term as follows (Anderson and van Wincoop, 2004): τ ij d γ ij eξb ij+ζe ij ςl ij ηc ij +ɛ ij (10) where d ij denotes the distance between country i and j; where b ij is a dummy variable taking the value 1, if the ow X ij takes place between a country belonging to a certain regional-bloc (like a monetary, economic or customary zone) and a country not belonging to this bloc, and 0 otherwise. 5 Where e ij takes value 1, if at least one of the two countries is landlocked, and 0 otherwise ( see Faye et al., 2004); l ij is a dummy variable taking value 1, if both partners have a common language 6 and 0 otherwise (Melitz, 2008); and c ij is 3 Doing so, we can control for a possible endogeneity of the two variables trade ows and gdp. See Emlinger et al. (2008). 4 The error term can enter the model in many ways. Here we introduce it via the trade cost τ ij. Doing so can be justied on the basis that trade costs are observed imperfectly. 5 This dummy is intended to estimate the border eect of dierent blocs in Africa, and is made more explicit below. 6 Here we consider each country's' rst ocial language only. 8

9 an another dummy that equals value 1, if the two countries have a common currency, and 0 otherwise (Frankel and Rose, 2002). The terms ɛ ij are assumed i.i.d error terms. Substituting equation (10) in equation (9) then yields the following equation: ln Z ij = σ ln L (σ 1)γ ln d ij (σ 1)ξ b ij (σ 1)ζẽ ij (σ 1)ς l ij (σ 1)η c ij (11) σ ln w i (σ 1) k L k L ln Z kj + ε ij where d ij d ij /Π k d L k/l kj are relative distances; b ij b ij L k k L b kj are relative borders. More precisely, equation (11) reects the trade resistance across blocs as compared to trade within blocs. ẽ ij e ij L k k L e kj capture the relative eects of countries being landlocked; l ij l ij L k k L l kj are the relative impacts of a common language use; and c ij c ij L k k L c kj are the relative eects of a common currency use. As regards the error structure, there many ways of modelling the error structure about which theory little to say. Behrens et al. (2012) pointed out that he error terms exhibit some form of cross-sectional correlation as: ε ij = λ k L u kj + u ij, where u ij (σ 1)ɛ ij is an iid error term. Note that, from equation (11) all variables superscripted with a tilde are measured as deviations from their population weighted averages, that allowing us to implicitly control for multilateral resistance. Moreover, equation (11) allows us to capture a possible spatial autocorrelation in error terms. Since wages are unobserved for most countries, rather than taking gdp per capita as proxies (as in Redding and Venables, 2004) which is clearly an endogenous variable in particular when we include intra-trade ow X ii, we prefer to introduce origin and destination country xed eects following Rose and van Wincoop (2001) and Feenstra (2002). Moreover, using the xed eects approach allows us to control for possible omitted variables. That is, let δ 1i denote an indicator variable that is 1 if country i is the exporter, 7 and 0 otherwise; and let δ 2j denote an indicator variable that is 1 if country j is the importer, and 0 otherwise. Then our spatial econometric reduced form to be estimated is: L k ln Z ij = β 0 +β 1 ln d ij +β 2 bij +β 3 ẽ ij +β 4 lij +β 5 c ij +β 6i δ 1i +β 7j δ 2j +ρ k L k L ln Z kj+ε ij (12) where β 0 σ ln L < 0 is the constant term; β 1 (σ 1)γ < 0 is the distance coecient of deviation from population weighted averages distances. Indeed, dij is a proxy variable for natural trade resistance which in turn is a composite of transport cost, 7 The number of dummy variables relative to country xed eects introduced is one less than the number of exporters, so as to avoid perfect collinearity. 9

10 transport time, and economic horizon. Consequently, it is hypothesized to have a negative eect on bilateral trade ows; β 2 (σ 1)ξ < 0 is the coecient that captures the border eect. Since the border eect reects the trade resistance resistance across blocs, we except a negative impact for its coecient. β 3 < 0 is the coecient of landlocked countries; β 4 > 0 captures the impact of common language; β 5 > 0 is the coecient of common currency. β 6i are the coecients of origin xed eects; β 7j are the coecients of destination xed eects and ρ (σ 1) < 0 is the spatial autoregressive coecient. Since the trade ow X ij from country i to country j also depends on all the trade ows from the other countries k to country j, we dene the n 2 n 2 spatial interaction matrix, with W = [S diag(l)] I n where S is the n n matrix whose elements are all equal to 1; is the Kronecker product and diag(l) is dened as the n n diagonal matrix of the L k /L terms. 8 To be more explicit about W, let W diag = diag(l) I n denote the matrix containing only the diagonal elements of W. In accordance with Behrens et al. (2012), equation (12)can be rewritten in matrix form as follows: (I ρw diag )Z = β 0 I+β 1 d+β2 b+β3 ẽ+β 4 l+β5 c+β 6 δ 1 +β 7 δ 2 +ρ(w ρw diag )Z+(W ρw diag )ε. Since I ρw diag is, by construction, an invertible diagonal matrix, we can premultiply by its inverse to obtain the following expression: Z = β 0 I+β 1 d+β2 b+β3 ẽ+β 4 l+β5 c+β 6 δ 1 +β 7 δ 2 +ρ(w W diag )Z+(I ρw diag )ε. The elements between positions i n1 and (i + 1) of (I ρw diag ) 1, given by [ ] 1 + (σ 1) L i 1, L depend on the origin index i only which is xed and identical for all destinations. Therefore, the components of the transformed (overlined) vectors of coecients are given by: [ β 1i β (σ 1) L ] 1 [ i, β L 2i β (σ 1) L ] 1 [ i, β L 3i β (σ 1) L ] 1 i, L [ β 4i β (σ 1) L ] 1 [ i, β L 5i β (σ 1) L i L β 7i β 7 [ 1 + (σ 1) L i L ] 1, ρ i ρ i [ 1 + (σ 1) L i L ] 1 [, β 6i β (σ 1) L i L ] 1. We obtain a specication with a distinct set parameters for each country. The full model, therefore, has a 'club' structure since all parameters (including the spatial autoregressive ones) must be estimated locally for each country. Behrens et al. ] 1, (2012) refer to this model as the heterogeneous coecients model. Since one of our objectives is to assess 8 It is worth noting that the interaction matrix comes structurally from the theoretical model of Behrens et al. (2012). Elements of this matrix are dened by share of populations L k /L and not by some ad hoc denition of distance. 10

11 the impact of dierent regional agreements, we don't need to estimate all parameters (including the spatial autoregressive ones) locally for each country. Therefore, we constrain all coecients to be identical across countries, which Behrens et al. refer to as the homogeneous coecients model. In addition, in most studies on gravity model the coecients are supposed to be homogeneous across countries (Anderson and van Wincoop, 2003; 2004; Baier and Bergstrand, 2001; 2009). Constraining the coecients to be identical amounts to assuming that the diagonal elements of W are equal to zero in equation (12). In that case, the model simplies substantially and can readily be estimated using standard spatial econometric techniques. Moreover, in the spatial econometrics literature, an observation is not neighbour to itself by convention, so that the diagonal elements are zero (w ii = 0). (See Anselin (1988a) 2.5 Intra- and inter-bloc eects From equation (12) we decompose the border eect into two components: the tradeboosting intra-bloc eect and the trade-reducing inter-bloc eect of the border. 9 To disentangle the two components and to retrieve the full implied border eects (both intra-bloc and inter-bloc), we proceed as follows. First, we dene the border as the ratio of trade ows in a world with borders (Z ij ) to that which would prevail in a borderless world (Z ij ). Using (9) and (10), we then have: B ij Z ij = e θ[b L ij k k L b kj ] Z ij k ( Zij Z ij ) ρ L k L, (13) where the term e θ[b ij k L k L b kj ] subsumes the border frictions as a deviation from their population weighted average. Note that (13) denes a log-linear system of all the relative trade ows, which depend on all border eects. Let B stand for the n 2 1 vector of ln( Z ij ) and let b stand for n 2 1 vector of [b Z ij L k ij k L b kj ]. The log-linearized version of the system has the following solution, B = θ(i ρw) 1 b, which allows us to retrieve the border eect as the exponential of the foregoing expression. Note that (13) quite naturally depends upon where countries i and j are located. Four cases may therefore arise with respect to intra-bloc and inter-bloc trade. Let pop bloc L k k bloc L (resp., pop row L k k / bloc L ) stand for the regional-bloc (resp., the rest of the 9 Trade ows between a bloc member country and a non-member country. 11

12 world) population shares. It is readily veried that 10 ln B ij = θ pop row θ pop row θ pop bloc θ pop bloc if (i bloc and j bloc) if (i bloc and j / bloc) if (i / bloc and j bloc) if (i / bloc and j / bloc) (14) Equation(14) reveals several interesting points. First, the expressions for bloc-bloc and row-row can be interpreted as the trade-boosting eect generated by the presence of borders which increases trade ows within each bloc. The trade ows within each bloc will be larger in a world with borders than in a borderless world. The reason is that borders protect regional rms from competition and give them an advantage in the regional market. Second, the expressions for bloc-row and row-bloc can be interpreted as the trade-reducing eect of the border on trade ows across countries located in dierent blocs. The trade ows across blocs will be smaller in a world with borders than in a borderless world. Third, as in Anderson and van Wincoop (2003), smaller blocs will have larger implied border eects than large blocs since their magnitude depends positively on the size of the trading partner, as measured by its share of population. The reason is that the border aects smaller blocs more than it does larger blocs, as it creates trade frictions for a larger share of the total demand served by its rms. Finally, the full border eect (combining the trade-boosting and trade-reducing eects), is given by e 2θpop row countries belonging to the bloc and by e 2θpop bloc for countries not belonging to the bloc. 11 To measure the intensity of the border eect we are interested in the ve main African regional blocs: waemu, cemac, ecowas, comesa and sadc. The rst two are simultaneously preferential trade blocs and monetary unions with a common currency (the franc cfa). waemu and cemac each have their own single currency (with the same acronym, franc cfa) and/each of which is pegged to the euro. Although these two currencies are commonly referred to by the same name (franc cfa) and have the same value, they are not interchangeable or mutually convertible, so this is not one common currency bloc but two juxtaposed blocs (Abdih and Tsangarides, 2010). for 3 Empirical evidence 3.1 Data and econometrics Our sample contains 150 countries with pairs of trade ows. 37 of countries are African countries, 36 American countries, 34 Asian countries, 38 European countries and 10 See Behrens et al. (2012). 11 In this study, we focus only on countries belonging to the bloc. 12

13 5 Oceanic countries. The data set includes exports X ij (including internal absorption X ii ) between countries, gdps Y i and Y j of trading partners all measured in millions of us dollars for the year We then use cross-sectional data for the year We compute internal absorption as X ii gdp i j X ij. Trade ows are from the un comtrade database. 12 are obtained from the Penn World Table gdp, national currency and population (also in 2010) data The data set also contains bilateral distances (in kilometers) between capital cities and are from the cepii database. 14 They are computed using the great circle distance formula applied to the capitals' geographic coordinates. As regards the internal distances of the countries, we follow Redding and Venables (2004) by computing internal distances as d ii κ surface i /π. As estimation results are known to be somewhat sensitive to the measurement of internal distance (Head and Mayer, 2002) we use 1/3, 2/3 and 1 for κ. However, since our results are quite robust to these dierent values of κ, we report only for κ = 2/3. 15 and language also come from the cepii database. Landlocked position In our study, we constructed the theoretically implied interaction matrix W using the population share of each exporting country in our sample. 16 To deal with potential endogeneity of population shares, we use threeyear lagged values of this variable. Results are robust with respect to dierent lags. We also ran the regressions using the `current year' (i.e., 2010), and results were little sensitive (with no change at all in the qualitative results). That why we report only the results for the year Insofar as one of our objectives is to assess the impact of dierent regional agreements in Africa, we further have to deal with the well-known problem of zero trade ows. The zero values found in the trade database correspond in fact either to a genuine zero ow or to a ow below a certain reporting threshold. The latter are very low and are therefore assimilated to absence of trade. Consequently, a subset of the observations are believed to represent censored values, which result in a truncated distribution for the dependent variable observations. Since there is no generally agreed-upon method for doing so (Anderson and van Wincoop, 2004), we control for the potential zero ow outliers by including a dummy variable in all regression. Alternative methods have been used to control for zero trade ows like the Heckman procedure (such as Emlinger et al., 2008). However, these methods are not known to perform better or to be theoretically more sound (Felbermayr and Kohler, 2006). We therefore both use a zero dummy variable and the spatial tobit to evaluate the robustness relative to the methodology used. When spatial autocorrelation is present in econometric specication, ols is no longer Results using 1/3 and 1 for κ are available upon request. 16 The interaction matrix W is normalized by its eigenvalues. 13

14 appropriate: the estimators obtained by this method are not convergent if there is a lagged endogenous variable and they are inecient in the presence of spatial autocorrelation (Anselin, 1988a; Anselin and Bera, 1998). likelihood (Lee, 2004; LeSage and Pace, 2009). 17 The method widely used is the maximum In the next section, we perform several estimation procedures depending on the structure of error terms. First, we estimate the model (12) without spatial interdependence (i.e. ρ = 0) using ols. Secondly, we estimate two versions of the spatial autoregressive model (12): (i) a Spatial Autoregressive model (sar) where errors ε ij are assumed to be iid, (ii) a General Spatial Model (gsm) where errors have a spatial autoregressive structure ε ij = λ n 2 k i w kjε kj + u ij. In this last case, we approximate the moving average by a more general autoregressive error structure. Finally, in order to treat the zero trade ows issue, we use a zero dummy in ols, car and gsm specications and we use a spatial tobit (see LeSage and Pace, 2009 and see Xu and Lee, 2014, for a recent discussion of the asymptotic properties of the spatial tobit). 3.2 Results Our empirical model is the model (12) where we add dummies relative to the customs unions (cemac and waemu), free trade areas (sadc and ecowas) and a common market (comesa) to measure the trade-boosting eect and the trade-reducing eects. We also introduce some control variables as described below. Table 1 displays the full results: Table 1 around here The estimation of equation (12) shows that waemu and cemac are not signicant using ols and sar (Table 1 column 1 and 2). Distance negatively aects trade ows, suggesting that distant countries tend to trade less with each other. Distance is a proxy for transport costs and time so that a long transport time increases the costs of packaging perishable goods. We note that the theoretical model predicts that the spatial autocorrelation coecient ρ should be negative. This means that trade ow from i to j decreases with the value of sales X kj from any third country k into the destination market, because varieties are gross substitutes. Since spatial interdependence is captured by the spatial autoregressive coecient in our estimating equations, this coecient may be interpreted as a measure of spatial competition encapsulating both aspects of market power and 17 For estimation we used James LESAGE's Econometrics Toolbox which is available at 14

15 consumer preference for diversity. Moreover, since the estimate for the parameter ρ is signicantly dierent from zero, least-squares estimates are biased and inconsistent. In what follows we will focus on the results given by the spatial models (especially gsm and tobit models) that result from the theoretical model. As regards the border eects, the coecients associated with all relative borders are negative and signicant. These coecients allow us to capture relative trade resistance due to regional blocs. To assess the magnitudes of impacts arising from regional blocs, we turn to the summary measures of intra-bloc eects, inter-bloc eects and total impacts presented in Table 2. Table 2 around here The results suggest that regional integration substantially increases trade between waemu countries. Indeed, our results show that trade between waemu countries is 5.7 times higher than trade between waemu countries and non waemu countries. Furthermore, the trade-boosting intra-bloc coecient shows that the trade ows within waemu are 2.4 times larger in a world with borders (world with blocs) than in a world without borders (world without blocs). As regards the trade-reducing inter-bloc eect, we nd that the trade ows across waemu are times smaller in a world with borders than in a world without borders. Put dierently, the trade ows across the blocs experience the border eect, which has the consequence of reducing these trade ows. For cemac, the full border coecient indicates that trade ows within cemac are 6.4 times higher than trade ows across cemac. The creation of cemac boosts trade between member countries by 2.6 times and reduces trade ows with the rest of the world by a factor of 0.4. Our estimations for sadc show that the coecient of the full border eect is 3.2, the trade-boosting eect is 1.8 and the trade-reducing eect is 0.6. These results show that intra-sadc trade increased compared with non-sadc trade. Implementation of sadc led to an increase in intra-sadc and a reduction in trade with non-members. Note that South Africa was initially not in this bloc but now it constitutes a dominant member as in Africa as a whole. The region is therefore more dependent on South Africa as a source of imports than as a market for exports. For ecowas, the full border eect coecient is 2.9. The trade-boosting intra-ecowas eect coecient is 1.7 and the trade-reducing coecient is 0.6. These results indicate a small border eect for ecowas despite the presence of all waemu countries (see Table A) and powerful neighbors (Nigeria and Ghana). We nd that trade ows within comesa would be 4 times higher than trade ows across comesa. The creation of comesa boosts 15

16 trade between member countries by 2 and reduces trade ows with the rest of the world by a factor of 0.5. The coecient relative to the landlocked countries is negative and signicant, suggesting that landlocked countries lag behind their maritime neighbours in external trade. This can be attributed to distance from the coast and several aspects of dependence on transit neighbours such as neighbours' infrastructure, sound cross-border political relations, neighbours' peace and stability and neighbours' administrative practices. Landlocked countries not only face the challenge of distance, but also the challenges that result from dependence on passage through a foreign transit country, one through which trade from a landlocked country must pass in order to access international shipping markets. Landlocked countries are completely dependent on their transit neighbours' infrastructure to transport their goods to port. This infrastructure may be weak for many reasons, including lack of resources, mis-governance, conict and natural disasters. Weak infrastructures (ports, roads, rail) in African countries increase transport costs and are often agreater obstacle to trade than tari and non-tari barriers in importing countries. Given the magnitude of infrastructure needs in Africa, it is acknowledged that to correct these shortcomings, regional and continental solutions are required. We nd a positive and signicant coecient for common language. This nding points out that common language promotes bilateral trade by facilitating communication and easing transactions. Two individuals who speak the same language can communicate and trade with each other directly whereas those without a sucient knowledge of a common language must often rely on an intermediary or hire an interpreter. The additional complexity inherent in such a mediated relationship, the potential for costly errors and their increased cost may be large enough to prevent otherwise mutually benecial transactions from occurring. As regards the coecient of common currency, we also nd a positive and signicant coecient. This nding shows that common currency promotes bilateral trade by reducing the costs of international transactions. The intuition is that trade between areas that use a single currency is cheaper and easier than trade between areas with their own currencies. Finally, it follows from all foregoing that regional blocs (whether customs unions, free trade areas, monetary unions, etc.) broadly have a positive eect on intra-trade ows. Trade within regional blocs is increased whereas trade with non-member countries is reduced. We note that regional integration is more advanced in waemu and cemac than in other regional blocs. Furthermore, waemu and cemac are a major export markets for the dominant countries in the two blocs (Cameroon for cemac and Senegal, Benin and Côte d'ivoire for waemu). They are the prime export market for landlocked countries in both blocs. The small border eect for ecowas can be attributed to the 16

17 failure by the members of these blocs to reduce both tari and non-tari barriers to trade. We also note that the border eect is high for the blocs that are well advanced in their integration process, and it is small for the blocs lagging behind in their integration process. We conclude that the more advanced the integration process is, the more member countries tend to trade with each other and to reduce their imports and exports with third countries. 4 Conclusion In this paper we estimated the border eect by breaking it down into two components: the trade-boosting intra-bloc eect and the trade-reducing inter-bloc eect. To estimate both trade-boosting and trade-reducing eects we based our approach on Behrens et al. (2012) by deriving a gravity equation and taking into account spatial interdependence between trade ows. Doing so yields a spatial econometrics reduced form where we explore dierent specications of error terms or the treatment of zeros trade ows ( sar, gsm and spatial tobit). We nd that regional blocs (whether customs union, free trade area, monetary union, or whatever) have a positive eect on intra-trade ows. Regional blocs not only increase intra-trade ows but also reduce trade with other outside countries. We also note that the border eect is high for the blocs that are well advanced in their integration process, and it is small for the blocs lagging behind in their integration process. As regards the spatial eect, we nd that the spatial interdependence between trade ows is reected in a negative relationship. Moreover, we also control for common language, landlocked countries, and common currency and they are found to be important determinants as distance in explaining trade ows. Acknowledgements. We thank Kristian Behrens, Bruce Shearer, an anonymous referee as well as internal seminar participants in UQAM Department of Economics for their very helpful comments and suggestions. Earlier versions of this paper were presented at the 47th Annual Conference of the CEA, Montreal, Canada, May 30-June 2 (2013) and at the 53ème Congrès de la SCSE, Québec, Canada, mai (2013). We would like to thank all the participants at those congress and conference. Wilfried Koch acknowledges nancial support from FQRSC Québec (Grant NP ). Aligui Tientao gratefully acknowledges nancial support from Université de Bourgogne and Conseil régional de Bourgogne (N 2011 BQR O56). He would like to thank Julie Le Gallo, Daniel Mirza, Christophe Tavéra and Marie-Claude Pichery for their valuable comments and sugestions during his thesis defence. Aligui Tientao obviously does not forget to pay tribute to Cem 17

18 Ertur which allowed us to signicantly improve this paper by his remarks, comments and suggestions. Part of this research was done while Aligui Tientao was visiting uqam. The usual disclaimer applies. References [1] Abdih, Y. and C. Tsangarides (2010) `FEER for the CFA franc', Applied Economics, 42 (16): [2] Agbodji, A.E. (2007) `Intégration et Echanges Commerciaux Intra Sous-Régionaux: le cas de l'uemoa', Revue africaine de l'intégration, 1 (1): [3] Aitken, N.D. (1973) `The eect of the EEC and EFTA on European Trade: A Temporal Cross-Section Analysis', American Economic Review, 63 (5): [4] Anderson, J.E. (1979) `A Theoretical Foundation for the Gravity Equation', American Economic Review, 69 (1): [5] Anderson, J.E. and E. van Wincoop (2003) `Gravity with Gravitas: A Solution to the Border Puzzle', American Economic Review, 93 (1): [6] Anderson, J.E. and E. van Wincoop (2004) `Trade Costs', Journal of Economic Literature, 42 (3): [7] Anselin, L. (1988a) Spatial econometrics: Methods and models, Kluwer Academic Publishers: Dordrecht. [8] Anselin, L. (1988b) `A test for spatial autocorrelation in seemingly unrelated regressions', Economics Letters, 28 (4): [9] Anselin, L and A.K. Bera (1998) Spatial Dependence in Linear Regression Models with an Introduction to spatial Econometrics, in Handbook of Applied Economics Statistics, A. Ullah and D.E.A Giles, eds., New York: Marcel Dekker. [10] Baier, S.L. and J.H. Bergstrand (2001) `The Growth of World Trade: Taris, Transport Costs, and Income Similarity', Journal of International Economics, 53 (1): [11] Baier, S.L. and J.H. Bergstrand (2009) `Bonus vetus OLS: A simple method for approximating international trade-cost eects using the gravity equation', Journal of International Economics, 77 (1):

19 [12] Behar, A. and L. Cirera-i-Crivillé (2013) `Does it Matter Who You Sign With? Comparing the Impacts of North-South and South-South Trade Agreements on Bilateral Trade', Review of International Economics, 21: [13] Behar, A. and L. Edward (2011) `How Integrated is SADC? : Trends in Intra- Regional and Extra-Regional Trade ows and Policy', Policy Research Working Paper 5625, The World Bank, Development Research Group, Trade and Integration Team. [14] Behrens, K., C. Ertur and W. Koch (2012) `Dual Gravity: Using Spatial Econometrics to Control for Multilateral Resistance', Journal of Applied Econometrics, 27 (5): [15] Bergstrand, J.H. (1985) `The Gravity Equation in International Trade', Review of Economics and Statistics, 67 (3): [16] Bosker, M. and H. Garretsen (2012) `Economic Geography and Economic Development in Sub-Saharan Africa', The World Bank Economic Review : [17] Carrère, C. (2004) `African Regional Agreements: Their Impact on Trade with or without Currency Unions', Journal of African Economies, 13 (2): [18] Collier, P. and A.J. Venables (2009) `Commerce et Performance Economique: la Fragmentation de l'afrique importe-elle?', Revue d'économie du développement, 23 (4): [19] Coulibaly, S. and L. Fontagné (2006) `South-South Trade: Geography Matters', Journal of African Economies, 15 (2): [20] De Sousa, J. and J. Lochard (2012) `Trade and Colonial Status', Journal of African Economies, 21 (3): [21] Deardor, A.V. (1995) `Determinants of Bilateral Trade: Does Gravity Work in a Neoclassical World?', NBER Working Papers 5377, National Bureau of Economic Research, Inc. [22] Dixit, A.K. and J.E. Stiglitz (1977) `Monopolistic Competition and Optimum Product Diversity', American Economic Review, 67 (3): [23] Eaton, J. and S. Kortum (2002) `Technology, Geography, and Trade', Econometrica, 70 (5): [24] ECA (2008) `Assessing Regional Integration in Africa III: Towards Monetary and Financial Integration in Africa', UN Economic Commission for Africa. 19

20 [25] Emlinger, C., F. Jacquet and E. C. Lozza (2008) `Taris and other trade costs: assessing obstacles to Mediterranean countries' access to EU-15 fruit and vegetable markets', European Review of Agricultural Economics, 35 (4): [26] Evenett, S.J. and W. Keller (1998) `On Theories Explaining the Success of the Gravity Equation', NBER Working Papers 6529, National Bureau of Economic Research, Inc. [27] Faye, M. L., J. W. Mcarthur, J. D. Sachs and T. Snow (2004) `The Challenges Facing Landlocked Developing Countries', Journal of Human Development, 5 (1). [28] Feenstra, R.C. (2002) `Border Eects and the Gravity Equation: Consistent Methods for Estimation', Scottish Journal of Political Economy, 49 (5): [29] Felbermayr, G. J. and Kohler, W. (2006) `Exploring the Intensive and Extensive Margins of World Trade', Review of World Economics, 142 (4): [30] Foroutan, F. and L. Pritchett (1993) `Intra-Sub-Saharan African Trade: Is it too little?', Journal of African Economies, 2 (1): [31] Frankel, J. and A. Rose (2002) `An Estimate of the Eect of Common Currencies on Trade and Income', The Quarterly Journal of Economics, 17 (2): [32] Geda, A. and H. Kebret (2008) `Regional Economic Integration in Africa: A Review of Problems and Prospects with a Case Study of COMESA', Journal of African Economies, 17(3): [33] Goretti, M. and H. Weisfeld (2008) `Trade in the WAEMU: Developments and Reform Opportunities', IMF Working papers 08/68, International Monetary Fund. [34] Head, K and T. Mayer (2002) `Illusory Border Eects: Distance Mismeasurement Inates Estimates of Home Bias in Trade', Working Papers , CEPII research center. [35] Helpman, E (1984) `A Simple Theory of International Trade with Multinational Corporations', Journal of Political Economy, 92 (3): [36] Kelejian, H.H., G.S. Tavlas and P. Pavlos (2012) In the neighborhood: The trade eects of the Euro in a spatial framework. Regional Science & Urban Economics, 42 (1/2): [37] Krugman, P. (1980) `Scale Economies, Product Dierentiation, and the Pattern of Trade', American Economic Review, 70 (5):